Original Research Annals of Internal Medicine Opportunistic

advertisement
Original Research
Annals of Internal Medicine
Opportunistic Screening for Osteoporosis Using Abdominal Computed
Tomography Scans Obtained for Other Indications
Perry J. Pickhardt, MD; B. Dustin Pooler, MD; Travis Lauder, BS; Alejandro Muñ oz del Rio, PhD; Richard J. Bruce, MD; and Neil Binkley, MD
Background: Osteoporosis is a prevalent but underdiagnosed
condition.
Objective: To evaluate computed tomography (CT)– derived bone
mineral density (BMD) assessment compared with dual-energy
x-ray absorptiometry (DXA) measures for identifying osteoporosis
by using CT scans performed for other clinical indications.
Design: Cross-sectional study.
Setting: Single academic health center.
Patients: 1867 adults undergoing CT and DXA (n = 2067 pairs)
within a 6-month period over 10 years.
Measurements: CT-attenuation values (in Hounsfield units [HU])
of trabecular bone between the T12 and L5 vertebral levels, with
an emphasis on L1 measures (study test); DXA BMD measures
(reference standard). Sagittal CT images assessed for moderate-tosevere vertebral fractures.
Results: CT-attenuation values were significantly lower at all vertebral levels for patients with DXA-defined osteoporosis (P <
0.001). An L1 CT-attenuation threshold of 160 HU or less was
O
steoporosis is prevalent and treatable and conveys a
considerable lifetime fracture risk, yet it remains substantially underdiagnosed and undertreated (1– 4). Currently, nearly half of all female Medicare beneficiaries have
never undergone bone mineral density (BMD) testing (5),
and more than 80% of all persons with a major
osteoporosis-related fracture do not have BMD testing or
receive pharmacologic agents to reduce fracture risk (6).
Furthermore, because normal BMD and mild osteopenia
confer a very low risk for osteoporosis (7), efficient and
cost-effective stratification of the unscreened population
into groups at low and high risk for osteoporosis and fractures is desirable. Central dual-energy x-ray absorptiometry
(DXA) of the hips and lumbar spine is widely recognized as
the reference standard for diagnosing osteoporosis (8, 9),
but it is underutilized. Safe and cost-effective alternatives
to increase detection of this condition are needed.
More than 80 million computed tomography (CT)
scans were performed in the United States in 2011 (10),
most of which carry potentially useful information about
BMD. Retrieval of BMD data available on body CT ex-
90% sensitive and a threshold of 110 HU was more than 90%
specific for distinguishing osteoporosis from osteopenia and normal
BMD. Positive predictive values for osteoporosis were 68% or
greater at L1 CT-attenuation thresholds less than 100 HU; negative
predictive values were 99% at thresholds greater than 200 HU.
Among 119 patients with at least 1 moderate-to-severe vertebral
fracture, 62 (52.1%) had nonosteoporotic T-scores (DXA falsenegative results), and most (97%) had L1 or mean T12 to L5
vertebral attenuation of 145 HU or less. Similar performance was
seen at all vertebral levels. Intravenous contrast did not affect CT
performance.
Limitation: The potential benefits and costs of using the various
CT-attenuation thresholds identified were not formally assessed.
Conclusion: Abdominal CT images obtained for other reasons that
include the lumbar spine can be used to identify patients with
osteoporosis or normal BMD without additional radiation exposure
or cost.
Primary Funding Source: National Institutes of Health.
Ann Intern Med. 2013;158:588-595.
For author affiliations, see end of text.
www.annals.org
aminations ordered for other indications requires no additional cost, patient time, equipment, software, or radiation
exposure, and these data can be retrospectively acquired. It
could therefore expand population screening efforts for
osteoporosis.
In a recent feasibility study of adults who underwent
osteoporosis screening with DXA and colorectal cancer
screening with CT colonography, we showed that a single
CT measurement of vertebral attenuation was equivalent
to the more complex dedicated quantitative CT (QCT)
assessment but was considerably easier to obtain (10). The
purpose of this study was to evaluate CT-derived BMD
assessment compared with DXA screening by using CT
scans that were performed for other clinical indications in a
larger patient population, focusing on the L1 level because
it is easily identified as the first non–rib-bearing vertebra
and is included on all abdominal and thoracic CT scans in
routine practice.
METHODS
Patient Cohort
See also:
Print
Editorial comment...................................................630
The University of Wisconsin Health Sciences Institutional Review Board (Madison, Wisconsin) approved this
Health Insurance Portability and Accountability Act–
adherent study. For inclusion, patients had to have had
abdominal CT and central DXA scanning of the hips and
spine within 6 months. All imaging was done at our insti-
588 © 2013 American College of Physicians
Downloaded From: http://annals.org/ by a Chongqing University of Medical Sciences User on 04/21/2013
CT Screening for Osteoporosis
tution over 10 years ending in December 2009; image retrieval and study analyses were performed between 2010
and 2012.
Original Research
Context
Osteoporosis is underdiagnosed.
Dual-Energy X-Ray Absorptiometry
Contribution
Dual-energy x-ray absorptiometry of the lumbar spine
and proximal femora was performed using standard techniques on Lunar Prodigy densitometers (GE Healthcare,
Waukesha, Wisconsin). Patients were categorized as having
osteoporosis (T-score 三 2.5 or the presence of a
moderate-to-severe vertebral compression fracture), osteopenia (T-score between 1.0 and 2.5), or normal
BMD (T-score 三 1.0) by using the lowest reported Tscore (8, 11). Because low BMD at 1 site carries an
increased risk for fracture at other sites (12, 13), patients
are generally categorized and managed according to their
lowest central T-score. Furthermore, in a substantial subset
of cases, T-scores for 1 of the 2 central sites are not reported for various technical reasons. At least 1 valid reported T-score for the lumbar spine or hips was required
for study inclusion.
This study found that computed tomography (CT) scans
can be used for detecting vertebral osteoporosis by comparing CT scans obtained for other reasons with dualenergy x-ray absorptiometry (DXA) scans performed
within 6 months of the CT. Approximately half of patients
with CT-identified osteoporotic vertebral compression fractures had nonsteoporotic T-scores (DXA false-negative
results).
Computed Tomography
Abdominal CT was done using multidetector CT
scanners (LightSpeed Series, GE Healthcare) calibrated
daily to ensure accurate vertebral CT-attenuation numbers,
which reflect underlying BMD (Figure 1). We retrospectively accessed the CT images and evaluated vertebral
BMD on a standard radiology picture archiving and communication system workstation, with images viewed in soft
tissue and bone windows (windows define gray-scale assignment of the image display to emphasize particular tissues and do not influence attenuation or BMD values [Figure 1]) (14). We assessed vertebral BMD by placing a
single oval click-and-drag region of interest (ROI) over an
area of vertebral body trabecular bone and then measuring
CT attenuation in Hounsfield units (HU), with lower HU
(lower attenuation) representing less-dense bone, at each of
the T12 through L5 levels (Figures 1 and 2); this process is
identical to that used for measuring CT attenuation for
other clinical conditions (for example, adrenal adenomas,
renal lesion enhancement, and fatty liver assessment). We
avoided placing the ROI near areas that would distort the
BMD measurement (posterior venous plexus; focal heterogeneity or lesion, including compression fracture; and
imaging-related artifacts).
We assessed the presence of vertebral compression
fractures by using sagittal CT views of the lumbar spine
(Figure 2, B) by employing the Genant visual semiquantitative method (15), a widely accepted way of assessing vertebral fractures on conventional radiography that can be
easily applied to sagittal CT images. We counted only obvious moderate (grade 2, 25% to 40% loss of height) or
severe (grade 3, >40% loss of height) compression deformities to avoid ambiguity related to more subjective borderline or mild compression deformities. All potential
moderate-to-severe compression fractures identified on the
www.annals.org
Caution
Osteoporosis treatment is often initiated based on hip
fracture risk. The relationship between hip fracture risk
and bone mineral density (BMD) is stronger for hip than
for spine BMD measures.
Implication
CT scans obtained for other reasons can be used for
opportunistic osteoporosis screening without additional
radiation exposure or cost.
—The Editors
initial review were verified in a separate reading session for
final confirmation, further excluding any questionable mild
fractures.
Statistical Analysis
We used Kruskal–Wallis tests to compare CTattenuation values within BMD categories at each vertebral
level (T12 to L5). We constructed kernel-density plots of
L1 CT attenuation for normal, osteopenic, and osteoporotic groups by using a Gaussian kernel with bandwidth
selected according to the Silverman rule of thumb. We
calculated sensitivity and specificity, positive and negative
predictive values (PPVs and NPVs), and positive and negative likelihood ratios for CT imaging compared with
DXA imaging across the range of observed CT-attenuation
values at 5-HU increments to establish thresholds that
would yield high sensitivity (about 90%), high specificity
(about 90%), or a balance between the 2 for distinguishing
osteoporosis from nonosteoporosis (osteopenia and normal
BMD) and normal BMD from low BMD (osteoporosis
and osteopenia).
We calculated adjusted Wald (“approximate”) 95%
CIs for proportions (for example, sensitivity and specificity) (16) and based the 95% CIs for PPV and NPV on the
logarithmic method (17). We also report findings for more
extreme thresholds (<100 HU and >200 HU) intended
to further increase specificity for osteoporotic and normal
populations, and we report CT–DXA cross-classifications
at thresholds that divide the study sample approximately
evenly among BMD categories.
16 April 2013 Annals of Internal Medicine Volume 158 • Number 8 589
Downloaded From: http://annals.org/ by a Chongqing University of Medical Sciences User on 04/21/2013
Original Research
CT Screening for Osteoporosis
Figure 1. Trabecular L1 CT-attenuation values for BMD assessment on body CT scans.
Example of axial CT images at the L1 vertebral level in 4 patients (A through D) viewed in standard soft tissue (row 1) and bone (row 2) window settings.
Trabecular bone CT-attenuation values are shown in red for each oval region of interest; note that the attenuation measure (in HU) does not change
according to the CT window for viewing. The 4 patients represent sample BMDs ranging from low (osteoporosis) (A) to high (normal) (D), which is
more visually apparent on the soft tissue window setting (row 1). Assuming a study-derived CT-attenuation threshold for osteoporosis of 三145 HU (see
Results section for details), patient A has osteoporosis by both CT (attenuation value, 20 HU; L2 vertebral fracture [not shown]) and DXA (T-scores for
both lumbar spine and hip, 4.0). Patient B has osteoporosis by CT (attenuation value, 93 HU; severe L4 vertebral fracture [not shown]) and osteopenia
by DXA (lumbar spine T-score, 2.2; hip T-score, 1.6). Patient C has osteopenia by CT (attenuation value, 148 HU) and DXA (lowest T-score,
1.6). Patient D has normal BMD by CT (attenuation value, 210 HU) and DXA (lowest T-score, 0.1). BMD = bone mineral density; CT = computed
tomography; DXA = dual-energy x-ray absorptiometry.
We assessed CT performance across all thresholds for
L1 and all vertebral levels by using empirical receiveroperating characteristic (ROC) curve analysis (18). The
ROC analyses derive from univariate logistic regression
models, where DXA-based osteoporosis was the dependent
variable and CT attenuation the independent variable. We
also performed multivariable logistic regression by using all
6 single-level T12 to L5 attenuations as dependent variables; the linear predictor was then used to construct ROC
curves. We assessed areas under the ROC curve (AUCs)
and corresponding 95% confidence limits for each vertebral level and compared the values by using a nonparametric approach (19).
In additional analyses, we compared CT performance
with and without intravenous contrast administration (because venous enhancement could potentially elevate ROIattenuation values through volume-averaging effects).
Statistical calculations and graphics were performed
using R, version 2.12.2 (R Development Core Team,
Vienna, Austria) (20).
Role of the Funding Source
The National Institutes of Health funded this study.
The funding source had no role in study design, conduct,
or analysis or the decision to submit the manuscript for
publication.
RESULTS
The study comprised 2063 CT–DXA pairs in 1867
adults (1511 women [81%]; mean age, 59.2 years
[SD, 12.5]). The median time between abdominal CT and
DXA studies was 67 days (interquartile range, 27 to 118
days).
Abdominal CT was done for various clinical indications, most commonly for suspected mass or oncologic
work-up (n = 414); genitourinary (n = 402) or gastrointestinal (n = 398) reasons, including virtual colonoscopy;
and unexplained abdominal pain or symptoms (n = 374)
(Appendix Table 1, available at www.annals.org). Of the
2063 CT scans, 1126 (54.6%) were obtained after intravenous contrast administration; the remainder were unenhanced. Fewer than 10% of patients contributed more
than 1 CT–DXA pair because of imaging without and with
contrast; analyses restricted to 1 CT–DXA pair per patient
resulted in absolute changes to sensitivity or specificity less
than 1% for all thresholds.
The DXA screening reference standard categorized patients as osteoporotic (22.9%), osteopenic (44.8%), and
normal (32.3%); BMD categories at hip and vertebral levels differed in approximately half the study population
(Appendix Tables 2 and 3, available at www.annals.org).
Computed tomography–attenuation values differed signif-
590 16 April 2013 Annals of Internal Medicine Volume 158 • Number 8
Downloaded From: http://annals.org/ by a Chongqing University of Medical Sciences User on 04/21/2013
www.annals.org
CT Screening for Osteoporosis
icantly for the 3 DXA-defined BMD categories at all vertebral levels (P < 0.001) (Appendix Figure 1, available at
www.annals.org). Figure 3 shows the overlap of DXAdefined BMD categories by L1 CT attenuation; Figure 4 is
a scatterplot of L1 CT-attenuation values and DXA T-scores.
An L1 CT-attenuation threshold of 160 HU was 90%
sensitive and a threshold of 110 HU was more than 90%
specific for distinguishing osteoporosis from osteopenia
and normal BMD; a threshold of 135 HU resulted in a
balanced sensitivity and specificity of approximately 75%
for each (Table 1). A CT-attenuation threshold intended
to increase specificity for osteoporosis (<100 HU) yielded
a PPV for osteoporosis of 68.4%; most (82%) “falsepositive” cases were classified as osteopenic by DXA.
Appendix Table 4 (available at www.annals.org) shows
CT–DXA cross-classifications at thresholds that divide
the study sample approximately evenly among BMD
categories.
The AUC across CT thresholds at L1 to distinguish
osteoporosis from osteopenia or normal BMD (Figure 5)
Original Research
was 0.83 (95% CI, 0.81 to 0.85) without a significant
difference between intravenous contrast– enhanced (AUC,
0.84) and unenhanced (AUC, 0.83) CT scans (P = 0.91).
The AUCs were statistically similar in subsets of patients
stratified by year (2000 to 2003, 2004 to 2005, 2006 to
2007, and 2008 to 2010) to account for changes in diagnostic technology or procedures over the study period (data
not shown).
Table 2 shows the diagnostic accuracy of CT for distinguishing normal from abnormal BMD (osteopenia and
osteoporosis); an L1 CT-attenuation threshold of 135 HU
was approximately 90% sensitive, and a threshold of 190
HU was 90% specific. A CT-attenuation threshold intended to increase specificity for normal BMD (>200
HU) yielded an NPV for osteoporosis of 99%; Appendix
Figure 2 (available at www.annals.org) shows the change in
PPV and NPV for osteoporosis across CT thresholds from
75 to 200 HU. The AUC across thresholds at L1 to distinguish normal from abnormal BMD was 0.80 (CI, 0.78
to 0.82).
Figure 2. Opportunistic osteoporosis screening at abdominal CT in a 59-year-old woman undergoing colorectal cancer screening
(with CT colonography).
CT = computed tomography; DXA = dual-energy x-ray absorptiometry. A. Axial CT image at the L1 vertebral level viewed in a bone window setting
shows appropriate placement of the oval region of interest in the trabecular bone. The CT-attenuation value of 109 HU places this patient in the lowest
quintile, raising concern for osteoporosis. B. Sagittal CT view shows a moderate T12 compression fracture (arrow). Note that higher thoracic vertebral
bodies are also sometimes included on abdominal CT scans. C and D. DXA evaluation of the hips (C) and lumbar spine (D) performed 3 mo later
demonstrated osteopenic T-scores ranging from 1.1 to 1.9 (lowest T-score of 1.9 from L1 to L4 evaluation). Therefore, this represents a DXA falsenegative result.
www.annals.org
16 April 2013 Annals of Internal Medicine Volume 158 • Number 8 591
Downloaded From: http://annals.org/ by a Chongqing University of Medical Sciences User on 04/21/2013
CT Screening for Osteoporosis
Computed tomography performed similarly at other
(non-L1) vertebral levels (Appendix Table 5 and Appendix
Figure 3, available at www.annals.org); attenuation thresholds targeted to yield sensitivity and specificity of approximately 90% varied slightly at each level because of the
small average decrease in attenuation values toward the L3
level (Appendix Figure 1). The AUCs were similar using
single-level vertebral measurements, a single multilevel
(T12 to L5) average, and more complex model-based multilevel measures (Appendix Figure 3).
One hundred nineteen patients had at least 1 moderate or severe vertebral fracture (35 with fractures at multiple vertebral levels), 62 (52.1%) of whom had either osteopenic (n = 50) or normal (n = 12) DXA T-scores
(Figures 2 and 4). Presumed degenerative changes were
noted in DXA evaluation in most of these false-negative
cases, which may have resulted in spurious T-score results.
Vertebral level L1 attenuation (or mean vertebral attenuation if L1 itself was involved by fracture) was 145 HU or
less in 115 (96.6%) of these patients, compared with
39.0% of patients with a T-score greater than 2.5 and no
fracture. Figure 4 shows the generalized decrease in vertebral CT attenuation among patients with fractures, despite
a range of DXA T-scores encompassing normal values.
Figure 4. L1 CT-attenuation values and corresponding DXA
T-scores.
400
L1 CT Attenuation, HU
Original Research
300
200
100
0
–6
–5
–4
–3
–2
–1
0
1
2
3
4
DXA T-Score
Normal (T-score ≥ –1)
Mild osteopenia (–1.5 < T-score < –1)
Moderate osteopenia (–2 < T-score ≤ –1.5)
Advanced osteopenia (–2.5 < T-score ≤ –2)
Osteoporosis (T-score ≤ –2.5)
DISCUSSION
In this study of the diagnostic accuracy of a simple
BMD screening method for adults who have undergone
abdominal CT imaging for other clinical indications, we
report results suggesting that CT images could be used to
identify patients with osteoporosis or normal BMD. The
Figure 3. Distributions of L1 CT-attenuation values in
Compression fracture
The black circles represent patients with CT-detected T12 to L5 compression fractures (not shown here are 15 patients with L1 compression
fractures where reliable CT-attenuation measurement at this level was
not possible). Note the broad range of DXA T-scores among patients
with vertebral fracture, including normal scores. Overall, more than half
of all patients with fractures had a nonosteoporotic DXA T-score, but
97% had an L1 or mean vertebral attenuation 三145 HU. CT = computed tomography; DXA = dual-energy x-ray absorptiometry.
normal, osteopenic, and osteoporotic cohorts.
BMD category, by DXA
0.012
Normal (T-score ≥ –1) Osteopenia
(–2.5 < T-score < –1)
Relative Frequency
0.010
Osteoporosis (T-score ≤ –2.5)
0.008
0.006
0.004
0.002
0.000
0
100
200
300
400
500
L1 CT Attenuation, HU
Based on lowest central DXA T-score. BMD = bone mineral density; CT =
computed tomography; DXA = dual-energy x-ray absorptiometry.
method that we used requires a negligible amount of training and time; could be applied prospectively by the interpreting radiologist or retrospectively by a radiologist or
even nonradiologist; adds no cost; and requires no additional patient time, equipment, software, or radiation exposure. It should not be confused with QCT, which is
more labor-intensive; requires an imaging phantom or
angle-corrected ROI measurement of bone, muscle, and fat
at multiple levels (21); and involves additional money,
time, and radiation exposure. Current QCT-derived Tscores do not directly correspond to DXA T-scores, limiting QCT clinical utility (21). We previously showed that
the simpler approach used in this study was as or more
effective and more reproducible than spine QCT for predicting DXA results (14).
Although optimal implementation of this method of
CT screening for osteoporosis remains to be determined,
our data suggest ways that it could be used in practice,
depending on clinical objectives. Identifying persons with
very low BMD by CT (for example, <100 HU at L1)
592 16 April 2013 Annals of Internal Medicine Volume 158 • Number 8
Downloaded From: http://annals.org/ by a Chongqing University of Medical Sciences User on 04/21/2013
www.annals.org
Original Research
CT Screening for Osteoporosis
Table 1. Performance Characteristics of L1 CT-Attenuation Values for Distinguishing Osteoporosis From Nonosteoporosis in 2040 CT–
DXA Pairs*
Variable
Threshold Selected for Achieving
High Sensitivity (About 90%)
Threshold Selected for Achieving
High Specificity (About 90%)
Threshold Selected for Balanced
Sensitivity and Specificity
L1 CT attenuation, HU
三160
三110
三135
Sensitivity
TP/(TP + FN), n/N
Sensitivity (95% CI), %
415/461
90.0 (86.9–92.4)
240/461
52.1 (47.5–56.6)
348/461
75.5 (71.4–79.2)
Specificity
TN/(TN + FP), n/N
Specificity (95% CI), %
826/1579
52.3 (49.8–54.8)
1441/1579
91.3 (89.8–92.6)
1190/1579
75.4 (73.2–77.4)
PPV
TP/(TP + FP), n/N
PPV (95% CI), %
415/1168
35.5 (32.8–38.3)
240/378
63.5 (58.5–68.2)
348/737
47.2 (43.6–50.8)
NPV
TN/(TN + FN), n/N
NPV (95% CI), %
826/872
94.7 (93.0–96.0)
1441/1662
86.7 (85.0–88.2)
1190/1303
91.3 (89.7–92.7)
Positive likelihood ratio (95% CI)
1.89 (1.78–2.00)
5.96 (4.97–7.14)
3.06 (2.77–3.39)
Negative likelihood ratio (95% CI)
0.19 (0.14–0.25)
0.53 (0.48–0.58)
0.33 (0.28–0.38)
CT = computed tomography; DXA = dual-energy x-ray absorptiometry; FN = false-negative; FP = false-positive; NPV = negative predictive value; PPV = positive
predictive value; TN = true-negative; TP = true-positive.
* “Nonosteoporosis” is defined as having osteopenia or normal bone mineral density. In this table, osteopenia and normal bone mineral density (according to the DXA
T-score) are considered negative findings. Patients with moderate or severe compression fractures were categorized as osteoporotic. The total number of CT scans is fewer than
2063 because L1 attenuation could not be reliably measured in 23 cases.
www.annals.org
Figure 5. Receiver-operating characteristic curves for
predicting osteoporosis by using CT attenuation at L1.
1.0
0.8
0.6
Sensitivity
might allow for rapid identification of high-risk cohorts in
whom further evaluation or treatment is warranted. Determining persons with high-normal BMD by CT (for example, >200 HU at L1) might effectively exclude osteoporosis and would probably make DXA unnecessary.
For L1 trabecular CT-attenuation values between 100
and 200 HU, or for all patients across the range of attenuation values, various thresholds or ranges could be considered, perhaps on the basis of pretest probability or a
priori risk. For example, an L1 CT-attenuation threshold
of 160 HU (90% sensitive for distinguishing osteoporosis
from nonosteoporosis) may be suitable for high-risk cohorts where the aim is to minimize false-negative results;
a substantial subset of osteopenia cases would be included
in this population, leading to low test specificity. Alternatively, an L1 threshold of 110 HU (91% specific)
may be prudent for groups considered at lower risk to
minimize false-positive results yet still detect approximately
half (52%) of patients with osteoporosis. Other CTattenuation thresholds or vertebral levels can be used, depending on the population and screening objectives.
One clear advantage of CT over DXA BMD screening
is its ability to accurately identify unsuspected osteoporotic
compression fractures, which clearly diagnose osteoporosis
independent of the patient’s DXA T-score (1). Osteopenic
and normal DXA T-scores were prevalent among patients
with vertebral fractures in our study and in others (22, 23);
this finding highlights the limitations of DXA, particularly
in terms of BMD overestimation related to degenerative
0.4
0.2
All (AUC, 0.83 [95% CI, 0.81–0.85])
Contrast (AUC, 0.84 [95% CI, 0.81–0.87])
No contrast (AUC, 0.83 [95% CI, 0.79–0.86])
0. 0
0.0
0.2
0.4
0.6
0.8
1.0
1 – Specificity
Receiver-operating characteristic curves for the L1 vertebral level show
no statistically significant difference in AUC for CT scans performed
with and without intravenous contrast (P = 0.91). However, further
investigation is needed to determine whether any adjustment in specific
CT-attenuation thresholds is necessary. AUC = area under the receiveroperating characteristic curve; CT = computed tomography.
16 April 2013 Annals of Internal Medicine Volume 158 • Number 8 593
Downloaded From: http://annals.org/ by a Chongqing University of Medical Sciences User on 04/21/2013
Original Research
CT Screening for Osteoporosis
Table 2. Performance Characteristics of L1 CT Attenuation for Distinguishing Normal From Low Bone Mineral Density in 2040
CT–DXA Pairs*
Variable
Threshold Selected for Achieving
High Sensitivity (About 90%)
Threshold Selected for Achieving
High Specificity (About 90%)
Threshold Selected for Balanced
Sensitivity and Specificity
L1 CT attenuation, HU
三135
三190
三160
Sensitivity
TP/(TP + FN), n/N
Sensitivity (95% CI), %
549/613
89.6 (86.9–91.7)
284/613
46.3 (42.4–50.3)
453/613
73.9 (70.3–77.2)
Specificity
TN/(TN + FP), n/N
Specificity (95% CI), %
669/1427
46.9 (44.3–49.5)
1286/1427
90.1 (88.5–91.6)
1008/1427
70.6 (68.2–72.9)
PPV
TP/(TP + FP), n/N
PPV (95% CI), %
549/1307
42.0 (39.4–44.7)
284/425
66.8 (62.2–71.1)
453/872
51.9 (48.6–55.2)
NPV
TN/(TN + FN), n/N
NPV (95% CI), %
669/733
91.3 (89.0–93.1)
1286/1615
79.6 (77.6–81.5)
1008/1168
86.3 (84.2–88.2)
Positive likelihood ratio (95% CI)
1.69 (1.59–1.78)
4.69 (3.92–5.60)
2.52 (2.29–2.76)
Negative likelihood ratio (95% CI)
0.22 (0.18–0.28)
0.60 (0.55–0.64)
0.37 (0.32–0.42)
CT = computed tomography; DXA = dual-energy x-ray absorptiometry; FN = false-negative; FP = false-positive; NPV = negative predictive value; PPV = positive
predictive value; TN = true-negative; TP = true-positive.
* “Nonosteoporosis” is defined as having osteopenia or normal bone mineral density. In this table, osteopenia (according to the DXA T-score) is combined with osteoporosis;
normal bone mineral density is considered a positive result. Patients with moderate or severe compression fractures were categorized as having osteoporosis. The total number
of CT scans is fewer than 2063 because L1 attenuation could not be reliably measured in 23 cases.
changes (1, 24). Vertebral CT-attenuation values tended to be
low at levels other than the site of fractures in our study,
and the fractures themselves were readily visible on the
sagittal views. Our observations are consistent with prior
studies documenting that many patients without osteoporosis diagnosed by DXA will sustain fragility fractures (22,
23) and suggest that CT attenuation may be a more accurate risk indicator.
Another possible advantage of CT over DXA screening is its scalability. We are currently assessing automation
of CT BMD screening that would permit evaluation of
numerous patients within the picture archiving and communication system file storage. Future refinements in appropriate population-specific CT-attenuation thresholds
could be derived from large retrospective (or prospective)
cohorts. In the future, it may even be possible to incorporate CT-attenuation data into fracture risk assessment
tools, such as the FRAX tool (World Health Organization
Collaborating Centre for Metabolic Bone Diseases, Sheffield, United Kingdom) (25).
We emphasize L1 vertebral measures in this study for
several reasons. First, the results at L1 are as or more accurate than the results at other levels, including multilevel
assessment. The L1 level is easily identified, which improves efficiency and reproducibility. It is included on all
standard chest and abdominal CT scans, substantially increasing its potential for higher overall screening yield. Accuracy was unaffected by the presence or absence of intravenous contrast. Furthermore, the measurement can be
applied retrospectively, because CT scans are now typically
stored indefinitely in most electronic medical records (picture archiving and communication systems).
Our study has limitations. Fracture risk prediction and
osteoporosis treatment decisions are often determined by
DXA-based hip T-scores. Our analysis is based on vertebral
measures and excludes assessment of risk factors for clinical
fracture important to the FRAX tool and other fracture
risk calculators (26, 27). Computed tomographic evaluation of the hip for BMD assessment is much more complex
than the simple lumbar assessment that we propose, and
we are currently investigating the potential for deriving a
DXA-equivalent T-score for the hips from standard pelvic
CT scans by using a dedicated software tool.
Our analysis is also based on DXA measures as a reference standard for assessing BMD, but the patients with
vertebral fractures and osteopenia or normal BMD by
DXA in our study highlight the limitations of use of DXA
as a reference standard. This finding suggests that, in a
clinical setting, cases with nonosteoporotic T-scores but
very low CT-attenuation values may warrant further investigation for overlying degenerative changes. Finally, we did
not formally assess the potential benefits and costs of using
the CT thresholds that we identified, but we speculate that
increasing detection of osteoporosis, with subsequent appropriate treatment to reduce fracture risk, combined with
reducing the number of normal DXA studies, would be
expected to yield substantial health care cost savings (28).
594 16 April 2013 Annals of Internal Medicine Volume 158 • Number 8
Downloaded From: http://annals.org/ by a Chongqing University of Medical Sciences User on 04/21/2013
www.annals.org
CT Screening for Osteoporosis
In conclusion, we demonstrate how routine abdominal
CT scans obtained for other clinical indications can be
used for opportunistic osteoporosis screening without the
need for additional imaging, radiation exposure, cost,
equipment, or patient time. We also report accuracy statistics for various CT-attenuation thresholds, use of which
would vary depending on clinical and population screening
objectives. Refinement of our techniques and confirmation
of these findings might justify more routine reporting of
vertebral trabecular attenuation with readings of abdominal CT evaluations performed for any reason.
From the University of Wisconsin School of Medicine and Public
Health, Madison, Wisconsin.
Grant Support: By the National Institutes of Health (grants
1R01CA144835-01 and 1R01CA169331-01).
Potential Conflicts of Interest: Disclosures can be viewed at www
.acponline.org/authors/icmje/ConflictOfInterestForms.do?msNum
=M12-1803.
Reproducible Research Statement: Study protocol and data set: Not
available. Statistical code: Available from Dr. Pickhardt (e-mail,
ppickhardt2@uwhealth.org).
Requests for Single Reprints: Perry J. Pickhardt, MD, Department of
Radiology, University of Wisconsin School of Medicine and Public
Health, E3/311 Clinical Science Center, 600 Highland Avenue, Madison, WI 53792-3252; e-mail, ppickhardt2@uwhealth.org.
Current author addresses and author contributions are available at
www.annals.org.
References
1. Raisz LG. Clinical practice. Screening for osteoporosis. N Engl J Med. 2005;
353:164-71. [PMID: 16014886]
2. Ebeling PR. Clinical practice. Osteoporosis in men. N Engl J Med. 2008;358:
1474-82. [PMID: 18385499]
3. Bessette L, Ste-Marie LG, Jean S, Davison KS, Beaulieu M, Baranci M, et al.
The care gap in diagnosis and treatment of women with a fragility fracture.
Osteoporos Int. 2008;19:79-86. [PMID: 17641811]
4. Metge CJ, Leslie WD, Manness LJ, Yogendran M, Yuen CK, Kvern B;
Maximizing Osteoporosis Management in Manitoba Steering Committee.
Postfracture care for older women: gaps between optimal care and actual care.
Can Fam Physician. 2008;54:1270-6. [PMID: 18791104]
5. King AB, Fiorentino DM. Medicare payment cuts for osteoporosis testing
reduced use despite tests’ benefit in reducing fractures. Health Aff (Millwood).
2011;30:2362-70. [PMID: 22147865]
6. Leslie WD, Giangregorio LM, Yogendran M, Azimaee M, Morin S, Metge
C, et al. A population-based analysis of the post-fracture care gap 1996-2008: the
situation is not improving. Osteoporos Int. 2012;23:1623-9. [PMID: 21476038]
7. Gourlay ML, Fine JP, Preisser JS, May RC, Li C, Lui LY, et al; Study of
Osteoporotic Fractures Research Group. Bone-density testing interval and transition to osteoporosis in older women. N Engl J Med. 2012;366:225-33. [PMID:
22256806]
8. Lewiecki EM, Gordon CM, Baim S, Leonard MB, Bishop NJ, Bianchi ML,
et al. International Society for Clinical Densitometry 2007 Adult and Pediatric
Official Positions. Bone. 2008;43:1115-21. [PMID: 18793764]
www.annals.org
Original Research
9. Conry CM, Main DS, Miller RS, Iverson DC, Calonge BN. Factors influencing mammogram ordering at the time of the office visit. J Fam Pract. 1993;
37:356-60. [PMID: 8409889]
10. IMV. IMV 2011 CT Market Outlook Report. Des Plaines, IL: IMV Medical
Information Division; 2011.
11. National Osteoporosis Foundation. Clinician’s Guide to Prevention and
Treatment of Osteoporosis. Washington, DC: National Osteoporosis Foundation; 2010.
12. Baim S, Binkley N, Bilezikian JP, Kendler DL, Hans DB, Lewiecki EM,
et al. Official Positions of the International Society for Clinical Densitometry and
executive summary of the 2007 ISCD Position Development Conference. J Clin
Densitom. 2008;11:75-91. [PMID: 18442754]
13. Marshall D, Johnell O, Wedel H. Meta-analysis of how well measures of
bone mineral density predict occurrence of osteoporotic fractures. BMJ. 1996;
312:1254-9. [PMID: 8634613]
14. Pickhardt PJ, Lee LJ, del Rio AM, Lauder T, Bruce RJ, Summers RM,
et al. Simultaneous screening for osteoporosis at CT colonography: bone mineral
density assessment using MDCT attenuation techniques compared with the
DXA reference standard. J Bone Miner Res. 2011;26:2194-203. [PMID:
21590738]
15. Genant HK, Wu CY, van Kuijk C, Nevitt MC. Vertebral fracture assessment using a semiquantitative technique. J Bone Miner Res. 1993;8:1137-48.
[PMID: 8237484]
16. Agresti A, Coull B. Approximate is better than “exact” for interval estimation
of binomial proportions. The American Statistician. 1998;52:119-26.
17. Confidence interval and statistical guidelines. In: Altman D, Machin D,
Bryant T, MJ Gardiner, eds. Statistics with Confidence. 2nd ed. London: BMJ
Books; 2000:109-10.
18. Lumley T. Programmer’s niche: ROC curves. R News. 2004.
19. DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under
two or more correlated receiver operating characteristic curves: a nonparametric
approach. Biometrics. 1988;44:837-45. [PMID: 3203132]
20. R Development Core Team. R: A Language and Environment for Statistical
Computing. Vienna, Austria: R Foundation for Statistical Computing; 2012.
21. Engelke K, Adams JE, Armbrecht G, Augat P, Bogado CE, Bouxsein ML,
et al. Clinical use of quantitative computed tomography and peripheral quantitative computed tomography in the management of osteoporosis in adults: the
2007 ISCD Official Positions. J Clin Densitom. 2008;11:123-62. [PMID:
18442757]
22. Siris ES, Brenneman SK, Barrett-Connor E, Miller PD, Sajjan S, Berger
ML, et al. The effect of age and bone mineral density on the absolute, excess, and
relative risk of fracture in postmenopausal women aged 50-99: results from the
National Osteoporosis Risk Assessment (NORA). Osteoporos Int. 2006;17:56574. [PMID: 16392027]
23. Schuit SC, van der Klift M, Weel AE, de Laet CE, Burger H, Seeman E,
et al. Fracture incidence and association with bone mineral density in elderly men
and women: the Rotterdam Study. Bone. 2004;34:195-202. [PMID: 14751578]
24. Yu W, Glüer CC, Fuerst T, Grampp S, Li J, Lu Y, et al. Influence of
degenerative joint disease on spinal bone mineral measurements in postmenopausal women. Calcif Tissue Int. 1995;57:169-74. [PMID: 8574931]
25. Kanis JA, Johnell O, Oden A, Johansson H, McCloskey E. FRAX and the
assessment of fracture probability in men and women from the UK. Osteoporos
Int. 2008;19:385-97. [PMID: 18292978]
26. Hans DB, Kanis JA, Baim S, Bilezikian JP, Binkley N, Cauley JA, et al;
FRAX Position Development Conference Members. Joint Official Positions of
the International Society for Clinical Densitometry and International Osteoporosis Foundation on FRAX. Executive Summary of the 2010 Position Development Conference on Interpretation and use of FRAX in clinical practice. J Clin
Densitom. 2011;14:171-80. [PMID: 21810521]
27. Kanis JA, McCloskey EV, Johansson H, Oden A, Ström O, Borgström F.
Development and use of FRAX in osteoporosis. Osteoporos Int. 2010;21 Suppl
2:S407-13. [PMID: 20464374]
28. Nayak S, Roberts MS, Greenspan SL. Cost-effectiveness of different screening strategies for osteoporosis in postmenopausal women. Ann Intern Med. 2011;
155:751-61. [PMID: 22147714]
16 April 2013 Annals of Internal Medicine Volume 158 • Number 8 595
Downloaded From: http://annals.org/ by a Chongqing University of Medical Sciences User on 04/21/2013
Annals of Internal Medicine
Current Author Addresses: Drs. Pickhardt and Pooler, Mr. Lauder, and
Drs. Muñoz del Rio and Bruce: Department of Radiology, University of
Wisconsin School of Medicine and Public Health, E3/311 Clinical Science Center, 600 Highland Avenue, Madison, WI 53792-3252.
Dr. Binkley: Divisions of Geriatrics and Endocrinology, University of
Wisconsin School of Medicine and Public Health, 2870 University Avenue, Suite 100, Madison, WI 53705.
Author Contributions: Conception and design: P.J. Pickhardt, B.D.
Pooler, R.J. Bruce, N. Winkler.
Analysis and interpretation of the data: P.J. Pickhardt, B.D. Pooler, A.
Muñoz del Rio, R.J. Bruce, N. Winkler.
Drafting of the article: P.J. Pickhardt, B.D. Pooler, A. Muñoz del Rio.
Critical revision of the article for important intellectual content: P.J.
Pickhardt, B.D. Pooler, A. Muñoz del Rio, R.J. Bruce, N. Winkler.
Final approval of the article: P.J. Pickhardt, B.D. Pooler, A. Muñoz del
Rio, R.J. Bruce, N. Winkler.
Statistical expertise: B.D. Pooler, A. Muñoz del Rio.
Obtaining of funding: P.J. Pickhardt.
Administrative, technical, or logistic support: P.J. Pickhardt, R.J. Bruce.
Collection and assembly of data: P.J. Pickhardt, B.D. Pooler, T. Lauder,
R.J. Bruce.
Appendix Table 3. DXA T-Score Categorization According to
Vertebral or Hip Location*
Appendix Table 1. Indications for Abdominal CT Scans
Indication
CT Scans, n
Suspected focal abnormality/mass or oncologic surveillance
Genitourinary signs*
Gastrointestinal signs†
Nonspecific symptoms‡
Vascular signs§
Suspected infection, inflammation, or abscess
Other or unspecified
Total
414
402
398
374
181
154
140
2063
CT = computed tomography.
* For example, urolithiasis, hematuria, renal evaluation, and adrenal lesion.
† For example, virtual colonoscopy, bowel obstruction, and diverticulitis.
‡ For example, abdominal pain, fever, weight loss, and trauma.
§ For example, abdominal aortic aneurysm and suspected bleeding or ischemia.
Vertebral
DXA T-Score
(n = 2063)
Hip DXA T-Score (n = 2063)
Normal
Osteopenia
Osteoporosis
Not
Reported
Normal
Osteopenia
Osteoporosis
Not reported
535 (4)
131 (1)
4 (0)
81 (4)
302 (11)
408 (22)
116 (7)
173 (19)
23 (4)
85 (6)
121 (23)
61 (13)
5 (0)
9 (1)
8 (4)
0 (0)
DXA = dual-ener gy x-ray absorptiometry.
* Numbers in parentheses indic ate that a moderate or severe vertebr al fracture is
present.
Appendix Figure 1. CT-attenuation data relative to DXA
T-scores.
Appendix Table 2. BMD Categorization in 2063 CT Scans
250
With DXA Correlation
Osteoporosis
Osteopenia
Normal
Not reported
Patients With CT–DXA Pairs, n
Hip
DXA
(Alone)
Lumbar
Spine
DXA
(Alone)
Central DXA
(Hip and
Lumbar
Spine)*
Central
DXA and
Vertebral
Fracture†
249
586
913
315
290
948
803
22
418
978
667
0‡
480
928
655
0‡
Mean CT Attenuation, HU
BMD Category
230
210
190
170
150
130
110
90
70
CT = computed tomography; BMD = bone mineral density; DXA = dualenergy x-ray absorptiometry.
* Determined by the lowest DXA T-score between the hip and lumbar spine.
† The presence of a moderate or severe vertebral compression fracture results in
osteoporotic categorization. This column represents the final reference standard
used in this study.
‡ Cases without both hip and lumbar spine DXA T-scores were excluded from the
study cohort.
50
T12
L1
L2
L3
L4
L5
Average
Vertebral Level
Osteoporosis
Osteopenia
Normal
Mean trabecular CT-attenuation values (SDs) at each vertebral level,
stratified by osteoporosis, osteopenia, and normal bone mineral density
according to the DXA reference standard. The differences between mean
attenuation for each bone mineral density group at each level are significant (P < 0.001). On average, CT attenuation tends to be lowest at the
L3 level and increases slightly at higher and lower levels. CT = computed tomography; DXA = dual-energy x-ray absorptiometry.
W-240 16 April 2013 Annals of Internal Medicine Volume 158 • Number 8
Downloaded From: http://annals.org/ by a Chongqing University of Medical Sciences User on 04/21/2013
www.annals.org
Appendix Table 4. Sample L1 CT-Attenuation Ranges for
Normal, Intermediate, and Abnormal Values Relative to
DXA T-Scores in 2040 CT–DXA Pairs*
L1 CT-Attenuation
Value†
Normal
Intermediate
Abnormal
Hip DXA T-Score
Normal
Osteopenia
Osteoporosis
417
287
96
180
410
344
13
66
206
Not Reported
3
5
13
Lumbar Spine DXA T-Score
Normal
Intermediate
Abnormal
Normal
447
335
128
Osteopenia
98
277
208
Osteoporosis
7
56
180
Not Reported
61
100
143
Central (Combined) DXA T-Score‡
Normal
Intermediate
Abnormal
Normal
366
197
56
Osteopenia
228
467
314
Osteoporosis
19
104
289
Not Reported
0
0
0
CT = computed tomography; DXA = dual-energy x-ray absorptiometry.
* Sample ranges were obtaining by roughly divided the cohort into thirds, but
other cutoff values can be used. 613 CT scans were classified as “normal,” 768 as
“intermediate,” and 659 as “abnormal.”
† Values >175 HU were categorized as “normal,” those between 130 and 175 HU
were categorized as “intermediate,” and those <130 HU were categorized as
“abnormal.”
‡ Only T-scores, not vertebral compression fractures, were considered.
www.annals.org
16 April 2013 Annals of Internal Medicine Volume 158 • Number 8 W-241
Downloaded From: http://annals.org/ by a Chongqing University of Medical Sciences User on 04/21/2013
Appendix Figure 2. PPV and NPV, according to L1 CT-attenuation threshold.
100
90
80
PPV
70
Percentage
NPV
60
50
40
30
20
75 80 85 90 95 100 105 110 115 120 125 130 135 140 145 150 155 160 165 170 175 180 185 190 195 200
L1 CT-Attenuation Threshold, HU
The graph shows the PPV for osteoporosis for L1 CT-attenuation values at or below each threshold; the NPV for excluding osteoporosis refers to L1
CT-attenuation values at or above each threshold. The relatively high NPV throughout is driven partly by the lower overall prevalence of osteoporosis
(22.9%). CT = computed tomography; NPV = negative predictive value; PPV = positive predictive value.
W-242 16 April 2013 Annals of Internal Medicine Volume 158 • Number 8
Downloaded From: http://annals.org/ by a Chongqing University of Medical Sciences User on 04/21/2013
www.annals.org
www.annals.org
Vertebral
Level
T12
L1
L2
L3
L4
L5
T12–L5ave
High (About 90%) Sensitivity
High (About 90%) Specificity
Balanced Sensitivity and Specificity
CT-Attenuation
Threshold, HU
Sensitivity
(TP/[TP + FN]),
% (n/N)
Specificity
(TN/[TN + FP]),
% (n/N)
CT-Attenuation
Threshold, HU
Sensitivity
(TP/[TP + FN]),
% (n/N)
Specificity
(TN/[TN + FP]),
% (n/N)
CT-Attenuation
Threshold, HU
Sensitivity
(TP/[TP + FN]),
% (n/N)
Specificity
(TN/[TN + FP]),
% (n/N)
165
160
155
150
145
150
150
89.3
90.0
91.3
90.9
90.4
90.6
89.4
49.4
52.3
51.5
47.1
55.6
56.0
56.7
115
110
110
100
100
105
110
51.3
52.1
54.4
50.9
48.7
50.4
55.7
90.1
91.3
89.0
88.5
88.6
89.4
89.6
140
135
130
125
125
130
130
75.2
75.5
72.4
75.7
75.6
73.3
74.7
74.0
75.4
75.4
72.3
72.8
74.1
75.8
(407/456)
(415/461)
(430/471)
(427/470)
(423/468)
(404/446)
(428/479)
(770/1560)
(826/1579)
(812/1577)
(743/1576)
(863/1553)
(839/1497)
(897/1583)
(234/456)
(240/461)
(256/471)
(239/470)
(228/468)
(225/446)
(267/479)
(1406/1560)
(1441/1579)
(1403/1577
(1394/1576)
(1376/1553)
(1339/1497)
(1418/1583)
(343/456)
(348/461)
(341/471)
(356/470)
(354/468)
(327/446)
(358/479)
(1155/1560)
(1190/1579)
(1189/1577)
(1139/1576)
(1130/1553)
(1110/1497)
(1200/1583)
CT = computed tomography; FN = false-negative; FP = false-positive; T12–L5ave = mean of the T12 to L5 values; TN = true-negative; TP = true-positive.
* “Nonosteoporosis” is defined as having osteopenia or normal bone mineral density.
1.0
0.8
0.6
0.4
0.2
0.0
0.0
0.2
0.6
1 – Specificity
0.4
0.8
T12 (AUC, 0.82 [95% CI, 0.80–0.84])
L1 (AUC, 0.83 [95% CI, 0.81–0.85])
L2 (AUC, 0.83 [95% CI, 0.8–0.85])
L3 (AUC, 0.81 [95% CI, 0.79–0.84])
L4 (AUC, 0.82 [95% CI, 0.79–0.84])
L5 (AUC, 0.82 [95% CI, 0.8–0.85])
Average (AUC, 0.84 [95% CI, 0.82–0.86])
1.0
Appendix Figure 3. Receiver-operating characteristic curves
for predicting osteoporosis by using CT values (region of
interest method).
Sensitivity
Multivariate model (AUC, 0.83 [95% CI, 0.81–0.86])
The AUCs are similar at each individual vertebral level, using a multilevel T12 to L5 average, and in a multivariable model incorporating
measures from all levels simultaneously, with broad overlap of 95% CIs.
Osteopenia was considered a false-positive result for these calculations.
The total number of assessable CT measurements per level was 2016 for
T12, 2040 for L1, 2048 for L2, 2046 for L3, 2021 for L4, and 1943 for
L5. AUC = area under the receiver-operating characteristic curve; CT =
computed tomography.
16 April 2013 Annals of Internal Medicine Volume 158 • Number 8 W-243
Downloaded From: http://annals.org/ by a Chongqing University of Medical Sciences User on 04/21/2013
Appendix Table 5. Diagnostic Performance of Sample Vertebral CT-Attenuation Thresholds at Abdominal CT for Distinguishing Osteoporosis From Nonosteoporosis*
Download